Daniel T. Davis

Integration of neural networks and decision tree classifiers for automated cytology screening

A squamous intraepithelial lesion (SIL) detection algorithm has been developed to process conventional Pap smears yielding a superior result (J.S.-J. Lee et al., 1990). The authors compare the object classification performance in an automated cytology screener. It consists of a Sun workstation, a DataCube image processing system, and an automatic stage/optics/illumination system. The system allows automated screening of 10 slides unattended. The main functional modules of the SIL algorithm include: image segmentation, feature extraction, and object classification. The classifiers used include neural network classifiers, statistical binary decision tree classifiers, a hybrid classifier, and the integration of multiple classifiers in an attempt to further improve algorithm performance.<<ETX>>

Attentional focus training by boundary region data selection

An attempt is made to improve the classification performance of a trained multilayer perceptron. Using inversion to locate boundary points of the partially trained classification surfaces, the authors have defined boundary regions and selected those training data which fell within the boundary regions. Continuing the training with only the boundary region data, the authors improved classification performance by 6% in an automated cytological classification application.<<ETX>>

Solving inverse problems using Bayesian modeling to incorporate information sources

Inverse problems have been considered unmanageable because they are often ill-posed, i.e., the statement of the problem does not thoroughly constrain the solution space. The authors propose taking advantage of this lack of information by adding additional informative constraints to the problem solution using Bayesian methodology. Bayesian modeling gains much of its power from its ability to isolate and incorporate causal models as conditional probabilities. As causal models are accurately represented by forward models, the authors propose converting implicit functional models into data driven forward models represented by neural networks, to be used as engines in a Bayesian modeling setting. Satellite remote sensing problems afford numerous opportunities for inclusion of ground truth information, prior probabilities, noise distributions, and other informative constraints within a Bayesian probabilistic framework. The authors apply these methods to an artificial satellite remote sensing problem, comparing the performance to a previously published method of iterative inversion of neural networks.<<ETX>>

Expanding Gaussian kernels for multivariate conditional density estimation

We demonstrate fundamental problems with the standard use of Gaussian kernels (SGKs) for estimating f(m|x) from sparse training data (x/sup i/,m/sup i/). We develop a new method that overcomes these considerations using Gaussian kernels with expanding covariances (EGKs) combined through Bayesian analysis. In addition, we demonstrate that for a synthetic problem, EGKs perform better qualitatively and quantitatively with respect to the Kullback-Leibler criteria.

Inversion of Parameters for Semiarid Regions by a Neural Network

Microwave brightness temperatures obtained from a passive radiative transfer model were inverted through use of a neural network. The model was applicable to semi-arid regions and produced dual-polarized brightness temperatures for 6.6, 10.7 and 37 GHz frequencies. A range of temperatures was generated by varying three geophysical parameters over acceptable ranges - soil moisture, vegetation moisture, and soil temperature. A multilayered perceptron (MLP) neural network was trained with a subset of the generated temperatures, and the remaining temperatures were inverted using a back propagation method. In addition, several synthetic terrains were devised and inverted by the network under local constaints. All the inversions showed good agreement with the original geophysical parameters, falling within 5% of the actual value of the parameter range.

Improved network inversion technique for query learning application to automated cytology screening

An improved neural network inversion technique that scales the search vector in accordance with the geometry of the problem has been developed. It searches in the direction of the gradient with a vector whose size is inversely related to the size of the gradient. To avoid unlimited growth of the search vector where the gradient is small, an upper bound is set on the size of the search vector. The network was trained by backpropagation and the training was halted when the network produced no error on the training set, where the output was categorized by binary thresholding. The results show the superior performance of the improved method. The technique was applied to automated cytology screening. A set of 400 object feature vectors randomly selected from a large database of 1929 feature vectors served as the initial training data.<<ETX>>

Solving inverse problems by Bayesian neural network iterative inversion with ground truth incorporation

Neural networks have long been applied to inverse parameter retrieval problems. The literature documents a development from the use of neural networks as explicit inverses to neural network iterative inversion (NNII) and, finally, to Bayesian neural network iterative inversion (BNNII), which adds a Bayesian superstructure to NNII. Inverse problems have been often considered ill posed, i.e. the statement of the problem does not thoroughly constrain the solution space. BNNII takes advantage of this lack of information by adding additional informative constraints to the problem solution using Bayesian methodology. This paper extends BNNII, showing how ground truth information, information regarding the particular parameter contour under reconstruction, and information regarding the underlying physical process, can be seamlessly added to the problem solution. Remote sensing problems afford opportunities for inclusion of ground truth information, prior probabilities, noise distributions, and other informative constraints within a Bayesian probabilistic framework. We apply these Bayesian methods to a synthetic remote sensing problem, showing that the addition of ground truth information, which is naturally included through Bayesian modeling, provides a significant performance improvement.

Inversion of Snow Parameters by Neural Network with Iterative Inversion

The inversion of snow parameters from passive microwave remote sensing measurements is performed with a neural network trained with a dense media multiple scattering model. A constrained iterative inversion scheme is used. Inversion of four parameters has been performed from five brightness temperatures. The four parameters are: mean-grain size of ice particles in snow, snow density, snow temperature and snow depth. The five brightness temperatures are that of 19 GHz vertical polarization, 19 GHz horizontal polarization, 22 GHz vertical polarization, 37 GHz vertical polarization, and 37 GHz horizontal polarization which are available from SSMI satellites. Based on the neural network constrained iterative inversion algorithm, we have also performed synthetic mapping of the terrain. Retrieval of synthetic mapping has been achieved. The incorporation of ground truth information is also considered.

The conditional binomial test revisited for clinical trials

McNemars test is used to test the hypothesis that one treatment is better than another in a matched-pair design for binary outcomes. The conditional binomial test in such a matched-pair design is the exact McNemar test. However, in many clinical trials, one wants to establish equivalency between two treatments. We discuss how to use a conditional binomial test to establish equivalency between two treatments in a matched-pair design. Sample size and power determination for each conditional binomial test are calculated. Some statistical properties of the tests are analyzed through Monte Carlo simulation.

Solving inverse problems by Bayesian iterative inversion of a forward model with ground truth incorporation

Inverse problems have been often considered ill-posed, i.e., the statement of the problem does not thoroughly constrain the solution space. In this paper we take advantage of this lack of information by adding informative constraints to the problem solution using Bayesian methodology. Remote sensing problems afford opportunities for inclusion of ground truth information, prior probabilities, noise distributions, and other informative constraints within a Bayesian probabilistic framework. We apply Bayesian methods to a synthetic remote sensing problem, showing that the performance is superior to a previously published method of iterative inversion of neural networks. In addition, we show that the addition of ground truth information, naturally included through Bayesian modeling, provides a significant performance improvement.